'Blocks' object has no attribute 'backwardC' error when calling blocks.conj.backward

[JanKirchhofTU]

Say I do the following simple thing:

import fastmat as fm
import numpy as np

# Say I have some matrix real or complex
some_matrix = fm.Matrix(
    np.random.randn(2,2)
)
some_complex_matrix = fm.Matrix(
    np.random.randn(2,2) + 1j*np.random.randn(2,2)
)
# And i create a fastmat.Blocks out of it:
blocks = fm.Blocks([[some_matrix, some_matrix], [some_matrix, some_matrix]])
complex_blocks =\
    fm.Blocks([[some_complex_matrix, some_complex_matrix],
                     [some_complex_matrix, some_complex_matrix]]
                  )
# Works
random_calc = blocks.backward(
    np.random.randn(blocks.shape[0])
)
# Works
random_calc = complex_blocks.backward(
    np.random.randn(blocks.shape[0])
)
# Works
random_calc = blocks.conj.backward(
np.random.randn(blocks.shape[0])
)
# Leads to the above error:
random_calc = complex_blocks.conj.backward(
    np.random.randn(blocks.shape[0])
)

The exact error output is

    random_calc = complex_blocks.conj.backward(
  File "fastmat/Matrix.pyx", line 1889, in fastmat.Matrix.Matrix.backward
  File "fastmat/Matrix.pyx", line 1948, in fastmat.Matrix.Matrix.backward
  File "fastmat/Matrix.pyx", line 2340, in fastmat.Matrix.Conjugate._backwardC
AttributeError: 'fastmat.Blocks.Blocks' object has no attribute 'backwardC'

[ChristophWWagner]

After some code-golf this is the shortest example I could come up with that reproduces the error:

>>> x = np.random.randn(10) * 1j
>>> y = fm.Diag(x).conj.backward(x)

Some observations:

• Error only occurs for complex matrices and complex transformation vectors and only for classes that apply the direct inplace transformation "cycthonCall" style, such as Blocks and Diag
• both, .conj.forward and .conj.backward are affected
• .H.T.forward()/.H.T.backward() is affected while .T.H.forward()/ .T.H.backward() is not. This could be a work-aroung for the moment! (since .conj should yield the same as .T.H)
• Other transformations (.H, .T, .inverse, .pseudoInverse) are not affected

[ChristophWWagner]

From the current revelations around cython, it could be worth investigating if this error is also susceptible to the version different of cython 0.29.15 to 0.29.20 as suggested in #87

[ChristophWWagner]

Issue was found and fixed in deaa900. Closing.